Power system natural disaster early warning and response method and device based on real-time monitoring

Abstract

The invention provides an electric power system natural disaster early warning and response method based on real-time monitoring, and relates to the technical field of electric power early warning response, and the method comprises the following steps: data analysis and processing: carrying out the analysis and processing of real-time monitoring data in an electric power system through combining a long short-term memory network with a Bayesian network; generating a scheduling scheme: updating the state of the power system in real time and generating an optimal scheduling scheme based on a multi-source data fusion and optimization method combining a Kalman filter and a genetic algorithm; generating an emergency response strategy: simulating the influence of natural disasters on the power system in real time in the virtual model through a digital twin technology and a self-adaptive dynamic planning method, and providing the emergency response strategy; and anomaly detection and self-healing: carrying out anomaly detection and self-healing on the operation state of the power system by combining big data analysis with edge calculation.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power early warning response, and in particular to a method and device for early warning and responding to natural disasters in an electric power system based on real-time monitoring. Background Art

[0002] After entering the 21st century, the power industry has developed rapidly. The increasingly improved infrastructure construction, access to various intelligent equipment and advanced regulation and control technologies have significantly improved the reliability of the power system. In recent years, with the development of the economy and society, the construction of distribution networks has received widespread attention. Some urban power grids have developed into large-scale urban distribution networks with complete facilities, advanced technology and high power supply reliability. However, extreme events, represented by extreme climate, pose an increasingly severe threat to the power supply capacity of high-proportion new energy power systems.

[0003] The prior art has the following technical problems when used:

[0004] 1. The lag problem of real-time data collection and processing: When facing natural disasters, traditional power systems are often unable to respond in time due to the lag of data collection and processing;

[0005] 2. Dispatch optimization and system stability issues: When natural disasters occur, the operating status of the power system may fluctuate dramatically, making traditional dispatching methods difficult to cope with.

[0006] 3. The problem of decision-making efficiency in emergency response: The decision-making process of traditional emergency response plans often relies on manual judgment, which is inefficient and easily affected by human factors. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention provides a natural disaster early warning and response method for a power system based on real-time monitoring, which solves the following problems:

[0008] Problems of lag in real-time data collection and processing, scheduling optimization and system stability, and decision-making efficiency in emergency response.

[0009] To achieve the above objectives, the present invention is implemented by the following technical scheme: a natural disaster early warning and response method for a power system based on real-time monitoring, the method comprising the following steps:

[0010] Sp1. Data analysis and processing: The data analysis and processing utilizes a long short-term memory network (LSTM) combined with a Bayesian network to analyze and process real-time monitoring data in the power system;

[0011] Sp2. Generate a dispatching plan: The dispatching plan is generated based on a multi-source data fusion and optimization method combining a Kalman filter and a genetic algorithm to update the state of the power system in real time and generate an optimal dispatching plan;

[0012] Sp3. Generate emergency response strategy: The emergency response strategy is generated by simulating the impact of natural disasters on the power system in real time in a virtual model through digital twin technology and adaptive dynamic programming method, and providing an emergency response strategy;

[0013] Sp4. Anomaly detection and self-healing: The anomaly detection and self-healing use big data analysis combined with edge computing to perform anomaly detection and self-healing on the operating status of the power system. The big data analysis is used to extract important features from historical data and detect anomalies, and the edge computing is used to update the distributed AI model and respond to anomalies.

[0014] Preferably, the combined algorithm of the LSTM neural network and the Bayesian network comprises the following steps:

[0015] Sp1, receive and process multi-dimensional time series data from the power system through LSTM neural network, including voltage, current, and frequency parameters, and generate hidden states for each time step;

[0016] Sp2. In each time step, the hidden state generated by LSTM is passed as input to the Bayesian network, which calculates the risk of disaster occurrence based on historical disaster data and system state probability distribution;

[0017] Sp3, based on the disaster risk probability calculated by the Bayesian network, the state weight of the LSTM network is updated to form a dynamic and continuously optimized prediction model for generating disaster warning signals in real time:

[0018] LSTM neural network: Input data: time series input data is X:

[0019] X={x1,x2,…,x T}

[0020] where x T is the input at time t.

[0021] The hidden state update formula in LSTM is as follows:

[0022] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0023] f t =σ(W f ·[ht-1 ,x t ]+b f )

[0024] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0025]

[0026] h t =o t *tanh(C t )

[0027] i t 、f t , o t are the activation values ​​of the input gate, forget gate, and output gate, respectively, and C t is the unit state, h t is the output vector, W i , W f , W o , W C is the weight matrix, b i 、b f 、b o 、b C is bias;

[0028] Bayesian network: Node representation: Bayesian network, where X = {X1, X2, ..., X n} is a set of random variables, G = (V, E) represents the directed acyclic graph of the network, V is the node set, and E is the edge set. Joint probability distribution: The joint probability distribution represented by the Bayesian network is:

[0029]

[0030] Pa(X i ) represents node X i By combining the time series prediction capability of LSTM with the probabilistic reasoning capability of Bayesian network, time dependency and uncertainty are handled in the prediction of complex natural disasters.

[0031] Preferably, the combined algorithm of the Kalman filter and the genetic algorithm comprises the following steps:

[0032] Sp1, extracting power system state information from real-time data collected by multiple sensors through Kalman filter, and calculating the optimal estimate of each state variable during the filtering process;

[0033] Sp2. Use genetic algorithm to globally optimize the state estimation results output by Kalman filter. The genetic algorithm evaluates the impact of different scheduling schemes on the stability of the power system through the fitness function and selects the optimal scheme;

[0034] Sp3, returning the optimal dispatching scheme selected by the genetic algorithm to the Kalman filter to correct future state estimation, thereby achieving dynamic dispatching and optimization of the power system;

[0035] Kalman Filter: State update formula: The state of the system is x k , the observed value is z k , the state transfer matrix is ​​F k , the observation matrix is ​​H k , the state update equation is:

[0036] x k =F k x k-1 +B k u k +w k

[0037] z k =H k x k +vk

[0038] w k 、v k are process noise and measurement noise respectively, B k is the control input matrix, u k For control input.

[0039] Prediction Steps:

[0040]

[0041] Update steps:

[0042]

[0043] are the predicted state estimate and covariance, K k Q k , R k . are the process noise covariance and the measurement noise covariance;

[0044] Genetic Algorithm:

[0045] Initial population: A population of N individuals is randomly generated, each of which represents a power grid dispatching scheme and is encoded as a binary string.

[0046] Fitness function: The fitness value of each individual is calculated according to its performance. The fitness function is based on the robustness and energy efficiency of the power system in disasters.

[0047] Selection, crossover and mutation: Select the best performing individuals for crossover and mutation to generate the next generation of population. The crossover operation generates new individuals by exchanging some gene sequences, while the mutation operation introduces diversity by randomly changing some genes of individuals.

[0048] Preferably, the steps of combining the digital twin technology with the adaptive dynamic programming method include:

[0049] Sp1. Create a virtual simulation model of the power system and dynamically calibrate the virtual model through real-time monitoring data to accurately reflect the current system status;

[0050] Sp2, using adaptive dynamic programming methods to perform multiple rounds of strategy iterations in a virtual model to evaluate the effectiveness of different emergency response strategies in simulated natural disaster scenarios;

[0051] Sp3. Based on the results of strategy iteration, determine and deploy the optimal emergency response strategy to the actual power system to achieve rapid response and recovery when disasters occur;

[0052] Digital twin technology: virtual model construction: the state of the real power system is X real , the state of the digital twin model is X twin , the state of the model is updated based on the sensor The real-time data S(t) is obtained and the equation f(X twin ,S(t)) simulates the operating status of the system:

[0053] X twin (t+1)=f(X twin (t),S(t))

[0054] Real-time monitoring and simulation: During disaster warning, the digital twin can be used to simulate the impact of disasters on the system in real time and provide simulation results of different response strategies.

[0055] Adaptive Dynamic Programming: State Value Function: Defining the State Value Function

[0056] V(s) is the long-term return in state s and is recursively updated by the Bellman equation:

[0057]

[0058] R(s,a) is the immediate benefit obtained by taking action a in state s, γ is the discount factor, and P(s′|s,a) is the probability of transitioning from state s to state s′;

[0059] Strategy update: Optimize the emergency response strategy of the power system through strategy iteration or value iteration. Combine the virtual simulation capability of digital twins and the optimal control capability of adaptive dynamic programming to achieve real-time emergency response and optimized control of the power system in natural disasters.

[0060] Preferably, the step of combining big data analysis with edge computing technology includes:

[0061] Sp1. Through big data analysis technology, principal component analysis (PCA) and autoencoder are used to reduce the dimension of the historical operation data of the power system, extract key features and construct feature vectors;

[0062] Sp2, deploy lightweight AI models in edge computing devices to process feature vectors obtained from sensors in real time, and use isolation forest algorithms and support vector machines (SVM) to detect system anomalies;

[0063] Sp3. When an abnormality is detected, the pre-trained distributed AI model is called through the edge computing device to perform local self-healing operations, and an alarm message is sent to the central control system to achieve a multi-level response;

[0064] Data analysis and anomaly detection: Feature extraction: Extract important features from a large amount of historical data through principal component analysis (PCA) or autoencoder. Anomaly detection: Use the isolation forest algorithm (IsolationForest) to detect anomalies in the system. There are N samples:

[0065] X={x1,x2,…,x N}

[0066] The anomaly score s(x) of each sample is calculated by the length of its isolation path in the tree structure:

[0067]

[0068] E(h(x)) is the average path length of sample x in the isolation tree, and c(N) is the adjustment coefficient of the number of samples.

[0069] Edge computing and distributed AI: Distributed model update: Run a lightweight neural network model f_θ on each edge device, and implement distributed updates of model parameters θ through federated learning:

[0070]

[0071] L is the loss function, K is the number of edge devices, and is the learning rate. The distributed AI model is used to quickly respond to and repair detected anomalies. By combining the anomaly detection capability of big data η analysis with the distributed AI self-healing mechanism of edge computing, efficient monitoring and rapid recovery of power systems in natural disasters can be achieved.

[0072] Preferably, the self-healing mechanism is achieved by the following steps:

[0073] Sp1. Use distributed AI models to monitor the operating status of the power system in real time. When a system abnormality is detected, it quickly triggers self-healing operations, including switching power lines, starting backup power sources, and adjusting load distribution.

[0074] Sp2. The optimization of self-healing operations is achieved through a reinforcement learning algorithm. The reinforcement learning model continuously optimizes future self-healing strategies based on the results of historical self-healing operations to ensure that the system stability can be quickly restored under different natural disaster conditions.

[0075] Preferably, the power system natural disaster early warning and response device of the method comprises:

[0076] Data acquisition module, used to collect real-time data of the power system from various sensors;

[0077] Prediction analysis module, which is used to predict the impact of natural disasters on the power system based on the combined algorithm of LSTM and Bayesian network;

[0078] The state update and optimization module is used to update the power system state in real time and generate a dispatch plan based on a combination of Kalman filter and genetic algorithm;

[0079] The emergency response module is used to simulate disaster scenarios and generate emergency response strategies through a combination of digital twins and adaptive dynamic programming algorithms;

[0080] The anomaly detection and self-healing module is used to detect anomalies in the power system and perform self-healing through a combined algorithm of big data analysis and edge computing.

[0081] Preferably, the prediction analysis module:

[0082] Use a multi-layer LSTM neural network to process multi-dimensional time series data of the power system, including voltage, current, frequency, wind speed, rainfall, seismic wave environment and system parameters, and generate hidden states and predicted values ​​for multiple time steps;

[0083] Based on the Bayesian network, uncertainty analysis and probability reasoning are performed on the predicted values ​​generated by the above LSTM neural network, and the probability distribution of natural disasters is calculated by combining historical natural disaster data, geographical location characteristics, and weather forecast information;

[0084] The disaster probability value calculated by the Bayesian network is compared with the preset risk threshold. When the threshold is exceeded, a disaster warning signal is generated and sent. At the same time, the learning rate and weight distribution of the LSTM neural network are adjusted to improve the prediction accuracy and response speed.

[0085] Preferably, the state updating and optimization module:

[0086] The Kalman filter integrates multiple sensor data in real time, including the operating status parameters of the power system (voltage, current, power factor) and external environmental parameters (temperature, humidity, wind speed, seismic waves), and outputs the optimal estimate of the system status;

[0087] The genetic algorithm is used to perform global search and optimization on the state estimates generated by the Kalman filter. The genetic algorithm evaluates the effectiveness of different scheduling schemes based on a predefined fitness function, which takes into account system stability, energy efficiency, and disaster response time factors.

[0088] Based on the optimal dispatching plan selected by the genetic algorithm, the operating mode of the power system is dynamically adjusted, including load distribution, generator start and stop, line switching, and energy storage device call-up. The Kalman filter is used to provide real-time feedback and corrections on the execution of the dispatching plan to cope with system fluctuations during disasters.

[0089] Preferably, the anomaly detection and self-healing module:

[0090] Through big data analysis technology, PCA and tSNE dimensionality reduction algorithms are used to process the historical operation data and real-time monitoring data of the power system, extract key feature vectors representing the health status of the power system, and construct a high-dimensional feature space model;

[0091] Deploy lightweight AI models in edge computing devices, use the isolation forest algorithm and the anomaly detection algorithm based on support vector machine (SVM) to monitor the operating status of the power system in real time in the high-dimensional feature space model and identify potential system anomalies;

[0092] When an anomaly is detected, the reinforcement learning algorithm is used to make self-healing operation decisions in the distributed AI model, optimize the self-healing strategy, and perform the following operations: switch the faulty line, start the backup power supply, adjust the load distribution, isolate the faulty area, and upload the abnormal information to the central control system for further analysis and global scheduling decisions.

[0093] The present invention provides a natural disaster early warning and response method for a power system based on real-time monitoring. It has the following beneficial effects:

[0094] 1. The present invention uses a data acquisition module to obtain the operating status and environmental parameters of the power system in real time through a variety of sensors, and uses LSTM neural network and Bayesian network for real-time prediction and analysis, which effectively improves the timeliness and accuracy of data processing, ensuring that the system can timely warn and take corresponding measures. LSTM neural network can capture the long-term and short-term dependencies in time series data, while Bayesian network combines historical data and external information for uncertainty analysis and probabilistic reasoning. This combined algorithm greatly improves the accuracy and reliability of prediction, enabling the system to better cope with the challenges of natural disasters.

[0095] 2. The state update and optimization module of the present invention uses the Kalman filter to optimally estimate the system state and uses the genetic algorithm to globally optimize the scheduling scheme. It can dynamically adjust the system operation mode, improve the system stability and energy efficiency, and ensure the reliability of power supply during disasters.

[0096] 3. The emergency response module of the present invention utilizes digital twin technology and adaptive dynamic programming algorithm to generate efficient emergency strategies based on virtual simulation, ensuring the implementation of optimal emergency measures in the shortest time, significantly improving the efficiency and accuracy of emergency response.

[0097] 4. The anomaly detection and self-healing module of the present invention uses big data analysis technology and edge computing to monitor the operating status of the system in real time, uses AI models and anomaly detection algorithms to promptly detect potential anomalies, and optimizes self-healing strategies through reinforcement learning algorithms to quickly take repair measures. This greatly enhances the self-healing ability of the power system in disasters and reduces system power outages and damage caused by anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 is a flow chart of the natural disaster early warning and response method of the present invention;

[0099] Figure 2 This is the curve diagram of wind speed, voltage and current changes in typhoon environment;

[0100] Figure 3 This is the load variation curve under typhoon environment;

[0101] Figure 4 This is the state diagram of the power system under typhoon environment;

[0102] Figure 5 This is the protection diagram of the power system in a typhoon environment;

[0103] Figure 6 It is a diagram of intensity change and equipment damage degree under earthquake conditions;

[0104] Figure 7 The load change and damage diagram under earthquake conditions;

[0105] Figure 8 It is a load trend diagram of the power system after the earthquake.

[0106] Fig. 9 It is the temperature and load prediction change diagram under the temperature environment;

[0107] Fig.10 Load adjustment optimization diagram for temperature environment;

[0108] Fig.11 It is the curve diagram of power load management in high temperature environment;

[0109] Fig.12 It is a schematic diagram of the structure of the natural disaster early warning and response device of the present invention. DETAILED DESCRIPTION

[0110] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Specific embodiment 1

[0112] A natural disaster early warning and response method for a power system based on real-time monitoring, the method comprising the following steps:

[0113] Sp1. Use the long short-term memory network (LSTM) combined with the Bayesian network to analyze the real-time monitoring data in the power system and predict the impact of natural disasters on the power system. Among them, LSTM is used to process time series data, and the Bayesian network is used to process uncertainty and probabilistic reasoning. In the power system, time series data (such as voltage, current, frequency, etc.) often have high time correlation and nonlinear characteristics. As a recursive neural network, the long short-term memory network (LSTM) is particularly suitable for capturing long-term dependencies. The structure of LSTM includes input gate, forget gate and output gate, which are used to selectively remember and forget information. At each time step, the input data affects the update of the cell state through the input gate, the forget gate determines which information needs to be forgotten, and the output gate determines the output of the current time step. For the power system, LSTM receives multi-dimensional time series data and generates a series of feature vectors representing the state of the power system by gradually updating the hidden state. The Bayesian network uses these feature vectors and historical data to calculate the risk probability of the power system suffering from natural disasters. In the Bayesian network, each node represents a random variable (such as the key state of the power system or external environmental factors), and the edge represents the conditional dependency between these variables. Through maximum a posteriori probability estimation, the Bayesian network can provide early warning signals for specific disaster scenarios. The combination of LSTM and Bayesian network realizes the dynamic prediction of disaster risks. By updating the state weights of LSTM in real time, the model can be continuously optimized to improve the accuracy and timeliness of early warning. The combined algorithm of LSTM neural network and Bayesian network includes the following steps:

[0114] Sp1, receive and process multi-dimensional time series data from the power system through LSTM neural network, including voltage, current, and frequency parameters, and generate hidden states for each time step;

[0115] Sp2. In each time step, the hidden state generated by LSTM is passed as input to the Bayesian network, which calculates the risk of disaster occurrence based on historical disaster data and system state probability distribution;

[0116] Sp3, based on the disaster risk probability calculated by the Bayesian network, the state weight of the LSTM network is updated to form a dynamic and continuously optimized prediction model for generating disaster warning signals in real time:

[0117] LSTM neural network: Input data: time series input data is X:

[0118] X={x1,x2,…,x T}

[0119] where x T is the input at time t.

[0120] The hidden state update formula in LSTM is as follows:

[0121] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0122] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0123] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0124]

[0125] h t =o t *tanh(C t )

[0126] i t 、f t , o t are the activation values ​​of the input gate, forget gate, and output gate, respectively, and C t is the unit state, h t is the output vector, W i , W f , W o , W C is the weight matrix, b i 、b f 、b o 、b C is bias;

[0127] Bayesian network: Node representation: Bayesian network, where X = {X1, X2, ..., X n} is a set of random variables, G = (V, E) represents the directed acyclic graph of the network, V is the node set, and E is the edge set. Joint probability distribution: The joint probability distribution represented by the Bayesian network is:

[0128]

[0129] Pa(X i ) represents node X i By combining the time series prediction capability of LSTM with the probabilistic reasoning capability of Bayesian network, time dependency and uncertainty are handled in the prediction of complex natural disasters.

[0130] Sp2. A multi-source data fusion and optimization method based on the combination of Kalman filter and genetic algorithm is used to update the state of the power system in real time and generate the optimal dispatching scheme, wherein the Kalman filter is used to fuse sensor data and the genetic algorithm is used to optimize the dispatching scheme of the power system; the Kalman filter is an optimal recursive algorithm for estimating the state of a dynamic system, which can handle noise data and uncertainty in the power system. Specifically, the Kalman filter recursively estimates the optimal value of the state of the power system through a prediction-update cycle based on the physical model and measurement data of the system. In the prediction step, the Kalman filter uses the state estimate and state transfer matrix of the previous time step to predict the system state of the current time step and estimate the prediction error covariance matrix; in the update step, the state estimate and error covariance matrix are updated through the current measurement data and measurement error covariance matrix. At the same time, the genetic algorithm performs global optimization based on the Kalman filter. The genetic algorithm simulates the natural selection process, including selection, crossover and mutation operations, to iteratively optimize the dispatching strategy of the power system. By evaluating the fitness of each scheduling scheme (such as power load balance, system safety and economy), the genetic algorithm finally selects the optimal scheduling scheme and feeds it back to the Kalman filter to adjust the state estimation of the next time step. In this way, the combination of Kalman filter and genetic algorithm can realize the dynamic scheduling optimization of the power system, especially in the case of disasters, and can quickly adjust the system state to cope with emergencies. The combined algorithm of Kalman filter and genetic algorithm includes the following steps:

[0131] Sp1, extracting power system state information from real-time data collected by multiple sensors through Kalman filter, and calculating the optimal estimate of each state variable during the filtering process;

[0132] Sp2. Use genetic algorithm to globally optimize the state estimation results output by Kalman filter. The genetic algorithm evaluates the impact of different scheduling schemes on the stability of the power system through the fitness function and selects the optimal scheme;

[0133] Sp3, returning the optimal dispatching scheme selected by the genetic algorithm to the Kalman filter to correct future state estimation, thereby achieving dynamic dispatching and optimization of the power system;

[0134] Kalman Filter: State update formula: The state of the system is x k , the observed value is z k , the state transfer matrix is ​​F k , the observation matrix is ​​H k , the state update equation is:

[0135] x k =Fk x k-1 +B k u k +w k

[0136] z k =H k x k +v k

[0137] w k 、v k are process noise and measurement noise, respectively, and B k is the control input matrix, u k For control input.

[0138] Prediction Steps:

[0139]

[0140] Update steps:

[0141]

[0142] are the predicted state estimate and covariance, K k Q k , R k . are the process noise covariance and the measurement noise covariance;

[0143] Genetic Algorithm:

[0144] Initial population: A population of N individuals is randomly generated, each of which represents a power grid dispatching scheme and is encoded as a binary string.

[0145] Fitness function: The fitness value of each individual is calculated according to its performance. The fitness function is based on the robustness and energy efficiency of the power system in disasters.

[0146] Selection, crossover and mutation: Select the best performing individuals for crossover and mutation to generate the next generation of population. The crossover operation generates new individuals by exchanging some gene sequences, while the mutation operation introduces diversity by randomly changing some genes of individuals.

[0147] Sp3. Through digital twin technology and adaptive dynamic programming methods, the impact of natural disasters on the power system is simulated in real time in a virtual model, and an emergency response strategy is provided. Among them, digital twin technology is used to create a virtual model of the power system, and adaptive dynamic programming is used to optimize the emergency response strategy; digital twin technology provides a virtual simulation environment consistent with the physical system for the power system. Through the input of real-time data, the state of the virtual model can be dynamically adjusted to accurately reflect the current state of the physical system. This virtual model includes not only the infrastructure of the power system, but also various simulation conditions under natural disaster scenarios. Through virtual simulation, the system can pre-evaluate the response strategies under different disaster scenarios. The adaptive dynamic programming (ADP) method is used to optimize these strategies. ADP evaluates the performance of each strategy in different disaster scenarios in turn through a strategy iteration method. After each round of strategy evaluation, ADP adjusts the decision function to improve the overall effect of the strategy. Finally, through multiple rounds of iterations, ADP determines the response strategy that performs best in various possible disaster scenarios and deploys it to the actual system. This method ensures that the power system can respond quickly when a disaster occurs and minimize losses. The steps of combining the digital twin technology with the adaptive dynamic programming method include:

[0148] Sp1. Create a virtual simulation model of the power system and dynamically calibrate the virtual model through real-time monitoring data to accurately reflect the current system status;

[0149] Sp2, using adaptive dynamic programming methods to perform multiple rounds of strategy iterations in a virtual model to evaluate the effectiveness of different emergency response strategies in simulated natural disaster scenarios;

[0150] Sp3. Based on the results of strategy iteration, determine and deploy the optimal emergency response strategy to the actual power system to achieve rapid response and recovery when disasters occur;

[0151] Digital twin technology: virtual model construction: the state of the real power system is X real , the state of the digital twin model is X twin , the state of the model is updated based on the sensor The real-time data S(t) is obtained and the equation f(X twin ,S(t)) simulates the operating status of the system:

[0152] X twin (t+1)=f(X twin (t),S(t))

[0153] Real-time monitoring and simulation: During disaster warning, the digital twin can be used to simulate the impact of disasters on the system in real time and provide simulation results of different response strategies.

[0154] Adaptive Dynamic Programming: State Value Function: Defining the State Value Function

[0155] V(s) is the long-term return in state s and is recursively updated by the Bellman equation:

[0156]

[0157] R(s,a) is the immediate benefit obtained by taking action a in state s, γ is the discount factor, and P(s′|s,a) is the probability of transitioning from state s to state s′;

[0158] Strategy update: Optimize the emergency response strategy of the power system through strategy iteration or value iteration. Combine the virtual simulation capability of digital twins and the optimal control capability of adaptive dynamic programming to achieve real-time emergency response and optimized control of the power system in natural disasters.

[0159] Sp4. Use big data analysis combined with edge computing technology to detect anomalies and self-heal the operating status of the power system. Big data analysis is used to extract important features from historical data and detect anomalies, and edge computing is used to update and respond to anomalies in distributed AI models. Anomaly detection and self-healing operations in power systems rely on the analysis of a large amount of historical data and real-time data. First, historical data is reduced in dimension by principal component analysis (PCA) or autoencoder, key features are extracted, and feature vectors are constructed. PCA projects the original data into a new coordinate system through linear transformation, so that these feature vectors retain the variance information of the data to the greatest extent. Autoencoder is a nonlinear dimensionality reduction method that extracts key features of the data by constructing a neural network that reconstructs the input data. These feature vectors are then passed to the lightweight AI model in the edge computing device. Edge computing devices are deployed on edge nodes of the power system and can process local data in real time, reducing the delay in data transmission. The lightweight AI model uses the isolation forest algorithm or support vector machine (SVM) to detect anomalies in the system. The isolation forest algorithm constructs multiple random binary trees to evaluate the "degree of isolation" of samples and detect potential anomalies. The support vector machine distinguishes between normal and abnormal states by constructing the optimal classification plane. When the edge computing device detects an anomaly, it immediately triggers the self-healing operation. The self-healing operation is performed by a pre-trained distributed AI model, which continuously optimizes the self-healing strategy through a reinforcement learning algorithm based on historical self-healing effects to ensure that the power system can recover quickly under different disaster conditions. In addition, the edge computing device will also feed back the abnormal information and self-healing operations to the central control system for global monitoring and further optimization of the response capability of the entire power system. .

[0160] The steps of combining big data analysis with edge computing technology include:

[0161] Sp1. Through big data analysis technology, principal component analysis (PCA) and autoencoder are used to reduce the dimension of the historical operation data of the power system, extract key features and construct feature vectors;

[0162] Sp2, deploy lightweight AI models in edge computing devices to process feature vectors obtained from sensors in real time, and use isolation forest algorithms and support vector machines (SVM) to detect system anomalies;

[0163] Sp3. When an abnormality is detected, the pre-trained distributed AI model is called through the edge computing device to perform local self-healing operations, and an alarm message is sent to the central control system to achieve a multi-level response;

[0164] Data analysis and anomaly detection: Feature extraction: Extract important features from a large amount of historical data through principal component analysis (PCA) or autoencoder. Anomaly detection: Use the isolation forest algorithm (IsolationForest) to detect anomalies in the system. There are N samples:

[0165] X={x1,x2,…,x N}

[0166] The anomaly score s(x) of each sample is calculated by the length of its isolation path in the tree structure:

[0167]

[0168] E(h(x)) is the average path length of sample x in the isolation tree, and c(N) is the adjustment coefficient of the number of samples.

[0169] Edge computing and distributed AI: Distributed model update: Run a lightweight neural network model f_θ on each edge device, and implement distributed updates of model parameters θ through federated learning:

[0170]

[0171] L(·) is the loss function, K is the number of edge devices, and is the learning rate. The distributed AI model is used to quickly respond to and repair detected anomalies. By combining the anomaly detection capability of big data η analysis with the distributed AI self-healing mechanism of edge computing, efficient monitoring and rapid recovery of power systems in natural disasters can be achieved.

[0172] The self-healing mechanism is implemented by the following steps:

[0173] Sp1. Use distributed AI models to monitor the operating status of the power system in real time. When a system abnormality is detected, it quickly triggers self-healing operations, including switching power lines, starting backup power sources, and adjusting load distribution.

[0174] Sp2. The optimization of self-healing operations is achieved through a reinforcement learning algorithm. The reinforcement learning model continuously optimizes future self-healing strategies based on the results of historical self-healing operations to ensure that the system stability can be quickly restored under different natural disaster conditions. Specific embodiment 2

[0176] A device for a natural disaster early warning and response method for a power system based on real-time monitoring, comprising:

[0177] Data acquisition module, used to collect real-time data of the power system from various sensors;

[0178] Prediction analysis module, which is used to predict the impact of natural disasters on the power system based on the combined algorithm of LSTM and Bayesian network;

[0179] The state update and optimization module is used to update the power system state in real time and generate a dispatch plan based on a combination of Kalman filter and genetic algorithm;

[0180] The emergency response module is used to simulate disaster scenarios and generate emergency response strategies through a combination of digital twins and adaptive dynamic programming algorithms;

[0181] The anomaly detection and self-healing module is used to detect anomalies in the power system and perform self-healing through a combined algorithm of big data analysis and edge computing.

[0182] The hardware components of the data acquisition module are:

[0183] Sensor network: includes voltage sensor, current sensor, temperature and humidity sensor, wind speed sensor, rainfall sensor, seismic wave sensor, etc. These sensors are connected to the data acquisition terminal via RS485 or CAN bus.

[0184] Data acquisition terminal: A hardware device with multi-channel data acquisition function that supports real-time acquisition of sensor data and uploads it to the central control system via Modbus or TCP / IP protocol.

[0185] Wireless communication module: used to upload the collected data to the cloud or control center through wireless communication protocols (such as LoRa, 4G / 5G, Wi-Fi) to achieve remote monitoring and data transmission.

[0186] The hardware components of the prediction analysis module are:

[0187] High-performance computing server: used to deploy LSTM neural network and Bayesian network models, with GPU acceleration function to improve computing efficiency and response speed.

[0188] Data storage device: A large-capacity storage device used to store historical data, training models, and real-time prediction results, supporting high-concurrency data read and write operations.

[0189] Network switches: used to connect computing servers and data storage devices to ensure low latency and high bandwidth for data transmission.

[0190] The hardware components of the status update and optimization module are:

[0191] Real-time controller: includes an industrial control computer with powerful computing power, supporting embedded operating system, capable of executing Kalman filter and genetic algorithm in real time.

[0192] Interface board: Equipped with a variety of communication interfaces (such as Ethernet, RS485, CAN), used to connect various sensors and actuators, collect power system status parameters, and send instructions.

[0193] Control Panel: A human-machine interface with a touch screen that allows operators to monitor system status, adjust parameters, and manually intervene in scheduling plans.

[0194] The hardware components of the emergency response module are:

[0195] Virtual simulation platform: Based on GPU-accelerated computing servers, digital twin technology is deployed to simulate power system operation scenarios and the impact of natural disasters, supporting large-scale parallel computing.

[0196] Adaptive dynamic programming hardware accelerator: A dedicated hardware accelerator (such as FPGA) is used to accelerate the calculation of the adaptive dynamic programming algorithm and improve the speed of emergency response strategy generation.

[0197] Communication network: Highly reliable industrial-grade Ethernet or wireless network for real-time transmission of simulation data and emergency strategies to the control center or distributed devices.

[0198] The hardware components of the anomaly detection and self-healing module are:

[0199] Edge computing equipment: edge computing nodes with high performance and low power consumption, equipped with AI inference acceleration chips, deploy lightweight AI models, and monitor the operating status of the power system in real time.

[0200] Data acquisition and processing unit: A processing unit equipped with algorithms such as PCA and tSNE for dimensionality reduction and feature extraction, and real-time analysis of the health status of the power system.

[0201] Self-healing control unit: It consists of an embedded controller and an actuator, which executes self-healing strategies in real time, switching faulty lines, starting backup power supplies, adjusting load distribution, and other operations.

[0202] The prediction analysis module:

[0203] Use a multi-layer LSTM neural network to process multi-dimensional time series data of the power system, including voltage, current, frequency, wind speed, rainfall, seismic wave environment and system parameters, and generate hidden states and predicted values ​​for multiple time steps;

[0204] Based on the Bayesian network, uncertainty analysis and probability reasoning are performed on the predicted values ​​generated by the above LSTM neural network, and the probability distribution of natural disasters is calculated by combining historical natural disaster data, geographical location characteristics, and weather forecast information;

[0205] The disaster probability value calculated by the Bayesian network is compared with the preset risk threshold. When the threshold is exceeded, a disaster warning signal is generated and sent. At the same time, the learning rate and weight distribution of the LSTM neural network are adjusted to improve the prediction accuracy and response speed.

[0206] The state update and optimization module:

[0207] The Kalman filter integrates multiple sensor data in real time, including the operating status parameters of the power system (voltage, current, power factor) and external environmental parameters (temperature, humidity, wind speed, seismic waves), and outputs the optimal estimate of the system status;

[0208] The genetic algorithm is used to perform global search and optimization on the state estimates generated by the Kalman filter. The genetic algorithm evaluates the effectiveness of different scheduling schemes based on a predefined fitness function, which takes into account system stability, energy efficiency, and disaster response time factors.

[0209] Based on the optimal dispatching plan selected by the genetic algorithm, the operating mode of the power system is dynamically adjusted, including load distribution, generator start and stop, line switching, and energy storage device call-up. The Kalman filter is used to provide real-time feedback and corrections on the execution of the dispatching plan to cope with system fluctuations during disasters.

[0210] The anomaly detection and self-healing module:

[0211] Through big data analysis technology, PCA and tSNE dimensionality reduction algorithms are used to process the historical operation data and real-time monitoring data of the power system, extract key feature vectors representing the health status of the power system, and construct a high-dimensional feature space model;

[0212] Deploy lightweight AI models in edge computing devices, use the isolation forest algorithm and the anomaly detection algorithm based on support vector machine (SVM) to monitor the operating status of the power system in real time in the high-dimensional feature space model and identify potential system anomalies;

[0213] When an anomaly is detected, the reinforcement learning algorithm is used to make self-healing operation decisions in the distributed AI model, optimize the self-healing strategy, and perform the following operations: switch the faulty line, start the backup power supply, adjust the load distribution, isolate the faulty area, and upload the abnormal information to the central control system for further analysis and global scheduling decisions.

[0214] In the workflow of the natural disaster warning and response device, the data acquisition module first monitors the system operation status and external environmental parameters in real time through the sensor network deployed at each key node of the power system. These sensors include voltage sensors, current sensors, temperature and humidity sensors, wind speed sensors, rainfall sensors, and seismic wave sensors, etc. All collected data will be collected in the data acquisition terminal. The data acquisition terminal has powerful data processing and transmission capabilities, and can upload the collected real-time data to the central control system through RS485, CAN bus or wireless communication module, and store it in the data storage device for further analysis.

[0215] After receiving the real-time data, the prediction and analysis module in the central control system first uses a multi-layer LSTM neural network to process these multi-dimensional time series data. The LSTM network can capture the complex time dependencies in the data and generate hidden states and prediction values ​​for multiple time steps. These prediction values ​​are then passed to the Bayesian network, which combines historical natural disaster data, geographic location characteristics, and weather forecast information for uncertainty analysis and probabilistic reasoning to calculate the probability distribution of natural disasters. When the calculated disaster probability value exceeds the preset risk threshold, the system automatically generates and sends a disaster warning signal, and adjusts the learning rate and weight distribution of the LSTM neural network to continuously improve the prediction accuracy and response speed.

[0216] During a disaster, the state update and optimization module acquires sensor data in real time, uses the Kalman filter to optimally estimate the operating state of the power system, and uses the genetic algorithm to globally search and optimize these state estimates. The genetic algorithm evaluates the effectiveness of different scheduling schemes based on a predefined fitness function, taking into account factors such as system stability, energy efficiency, and disaster response time. Ultimately, the system selects the optimal scheduling scheme and dynamically adjusts the operating mode of the power system, such as load distribution, generator start and stop, line switching, and energy storage device call. The module also uses the Kalman filter to provide real-time feedback and corrections on the execution of the scheduling scheme to ensure the stable operation of the power system during disasters.

[0217] When a disaster occurs, the emergency response module uses digital twin technology to simulate the actual operation scenario of the power system, and uses an adaptive dynamic programming algorithm to simulate the generation process of emergency response strategies under different disaster scenarios. The simulation results and emergency strategies will be transmitted to the control center or distributed execution devices in real time through the communication network to ensure that emergency measures can be implemented in a timely and accurate manner. The execution of emergency strategies includes but is not limited to load adjustment, startup of backup power supplies, line switching, and protection operations of key equipment to ensure the rapid recovery and stable operation of the power system in disasters.

[0218] Finally, the anomaly detection and self-healing module monitors the operating status of the power system in real time through a lightweight AI model deployed in the edge computing device, and uses big data analysis technology and anomaly detection algorithms such as the isolation forest algorithm and support vector machine (SVM) to identify potential system anomalies in the high-dimensional feature space model. Once an anomaly is detected, the system will automatically trigger a self-healing operation decision, use a reinforcement learning algorithm to optimize the self-healing strategy, and perform self-healing operations including fault line switching, starting backup power, adjusting load distribution, and isolating fault areas. At the same time, the abnormal information will be uploaded to the central control system to provide a basis for further global scheduling decisions. Specific embodiment 3

[0220] like Figure 2-5 As shown, power system protection and restoration during typhoons: Coastal cities are affected by super typhoons, which are expected to bring strong winds and heavy rains, causing power transmission lines to collapse, substations to be damaged, and even large-scale power outages. In order to ensure the stable operation and rapid recovery of the city's power grid, the city's power company needs to take precautions in advance and respond quickly when the typhoon passes.

[0221] Prediction and risk assessment before a typhoon arrives: Power companies use long short-term memory networks (LSTM) combined with Bayesian networks to analyze real-time monitoring data in the power system. The LSTM network processes time series data such as voltage, current, and wind speed to predict the impact of typhoons on the system at each time node. The Bayesian network infers the risk probability of facilities such as transmission lines and substations based on historical typhoon data and currently monitored environmental conditions.

[0222] Through this combined algorithm, the system can generate a detailed risk assessment report 24 hours in advance, identifying the areas and facilities most vulnerable to typhoons. The dispatch center can issue early warnings and formulate specific response measures for these high-risk areas.

[0223] Real-time status monitoring and dispatch optimization during typhoons: When a typhoon hits, the Kalman filter combines data from multiple sensors in the power system (wind speed, voltage, and current sensors) to update the system's operating status in real time. The Kalman filter filters out measurement noise and provides the best estimate of the system state. Subsequently, the genetic algorithm globally optimizes the power system's dispatch plan based on these real-time states.

[0224] The genetic algorithm will take into account the impact of instantaneous strong winds and heavy rains brought by typhoons and dynamically adjust the load distribution of the power grid. It will reduce or temporarily stop the power supply to certain high-risk lines to avoid the expansion of faults and give priority to the supply of electricity to key facilities such as hospitals. This dynamic scheduling allows the power system to maintain stable operation even during typhoons.

[0225] Disaster impact simulation and emergency response: Using digital twin technology, the power company created a virtual model that is highly similar to the actual power system. During the typhoon, the virtual model received real-time monitoring data from sensors and made dynamic adjustments based on the data to accurately reflect the current system status.

[0226] Combined with the adaptive dynamic programming method, the virtual model evaluates and optimizes different emergency response strategies in real time when simulating the impact of natural disasters on the power system. The system simulates multiple emergency plans, predicts indicators such as system recovery time and equipment damage under different plans, and selects the best plan. Ultimately, these optimized emergency response strategies will be quickly deployed in the actual power system to ensure power supply security during typhoons.

[0227] Anomaly detection and self-healing mechanism: During the typhoon, big data analysis technology extracted key features from historical data and combined with edge computing technology to detect anomalies in the power system. The AI ​​model deployed on the edge device analyzes sensor data in real time and uses the isolation forest algorithm to identify anomalies, such as signs of overload on a certain line.

[0228] Once an abnormality is detected, the system will immediately trigger a self-healing operation, automatically switch to the backup line, start the backup power supply, or adjust the load distribution to prevent the power outage from expanding. At the same time, the edge computing device will also report the abnormal situation to the central control system to ensure that the dispatch center can respond quickly. In this way, even if the power system fails during a typhoon, it can quickly self-repair and maintain the continuity of power supply.

[0229] Through this comprehensive technical solution, the power company successfully predicted the risks before the typhoon arrived and took preventive measures in advance. During the typhoon, the system automatically carried out dynamic scheduling and real-time simulation, quickly responded to abnormal situations and performed self-healing operations, ultimately effectively ensuring the city's power supply and avoiding large-scale power outages and equipment damage. Specific embodiment 4

[0231] like Figure 6-8 Figure 1. Power system restoration after an earthquake: A severe earthquake hit the city, damaging multiple substations and disrupting transmission lines. The city's power company faced the challenge of quickly restoring power supply, especially to key facilities such as hospitals and command centers.

[0232] Post-earthquake status assessment and risk prediction: After an earthquake, the power company uses the LSTM network combined with the Bayesian network to analyze the real-time earthquake monitoring data and power system operation data. The LSTM network processes time series data such as seismic waves and equipment status to predict the impact of earthquake aftershocks on the power system. The Bayesian network infers the secondary damage risks of different parts of the power system based on historical earthquake data and current monitoring data.

[0233] The system quickly generates a risk assessment report, identifies further damaged equipment and lines, and helps power companies prioritize emergency repairs and prevent secondary disasters from expanding.

[0234] Real-time dispatch and optimization: The state of the power system after the earthquake is complex and changes dynamically. The power company updates the system's state estimate in real time by fusing data from multiple sensors (seismic waves, current sensors) through the Kalman filter. Combined with genetic algorithms, the system globally optimizes the dispatch plan and gives priority to restoring power to key facilities such as hospitals and command centers.

[0235] Based on the state estimation provided by the Kalman filter, the genetic algorithm evaluates the impact of different scheduling schemes on the overall recovery of the system and selects the optimal scheme. In the early period after the earthquake, the system will recommend temporarily shutting down certain damaged lines and concentrating resources to restore power to key facilities.

[0236] Virtual simulation and emergency recovery strategy: After the earthquake, the power company used digital twin technology to create a virtual model of the power system. The virtual model was dynamically updated based on real-time data to simulate the impact of different repair plans on the system. Through adaptive dynamic programming, the system repeatedly iterated in the virtual environment to evaluate and optimize the emergency recovery strategy.

[0237] The system ultimately determines the optimal repair sequence and emergency measures, giving priority to repairing high-voltage lines or starting backup substations, and quickly implements these strategies in the actual power system to ensure that power supply is restored in the shortest time possible.

[0238] Anomaly detection and self-healing: Big data analysis technology extracts key features of the power system after the earthquake, and the AI ​​model on the edge computing device monitors anomalies in the recovery process in real time. Once an anomaly, line overload or equipment failure is detected, the system will immediately trigger a self-healing operation and send an alarm message to the central control system.

[0239] Self-healing operations include adjusting load distribution, starting backup equipment, etc. to ensure that no secondary failures occur during the recovery process, thereby improving recovery efficiency and reducing power outage time.

[0240] Through this comprehensive technical solution, the power company quickly assessed the system status after the earthquake and restored power supply to key load areas as a priority. Throughout the restoration process, the system ensured the stable operation of the power system through real-time simulation and optimized scheduling, and responded to abnormal situations in a timely manner, ultimately restoring normal power supply to the city in a relatively short period of time. Specific embodiment 5

[0242] like Figure 9-11 As shown in Figure 1, power load management during extreme high temperatures, the city experienced continuous days of extreme high temperatures in the summer, which led to a large number of residents and commercial users using air conditioners, and the power load increased sharply. The city power company needs to deal with risks such as line overload and equipment failure to ensure the stable operation of the power system under high load.

[0243] Load forecasting and risk assessment during high temperatures: Using LSTM networks combined with Bayesian networks, power companies predict power load during high temperatures. LSTM networks process historical high temperature weather and current load data to predict load change trends in the next few days. Bayesian networks combine historical data with real-time monitoring data to assess system risks caused by load surges and overload conditions on certain lines.

[0244] The load forecast report generated by the system helps the dispatch center to identify overloaded lines and equipment in advance, formulate corresponding response strategies, and adjust load distribution or start backup power supplies in advance.

[0245] Real-time state monitoring and dispatch optimization: During high temperatures, the Kalman filter integrates multi-source sensor data (temperature, current sensors) of the power system in real time to update the system's state estimate. Combined with genetic algorithms, the system optimizes the power dispatch plan to avoid large-scale power outages caused by local overloads.

[0246] Based on the state estimation provided by the Kalman filter, the genetic algorithm recommends adjusting the load by region, regulating the air conditioning load or starting the backup generator to share the power pressure and ensure the smooth operation of the system.

[0247] Emergency response and simulation evaluation: In the case of extreme high temperatures, digital twin technology is used to create a virtual model corresponding to the real power system. Combined with adaptive dynamic programming, the system repeatedly simulates different emergency response strategies in a virtual environment to evaluate their impact on the stability of the power system.

[0248] Based on the simulation results, the system determines the optimal load regulation strategy and emergency response measures and implements these strategies in the actual power system to ensure stable operation under high load conditions.

[0249] Anomaly detection and self-healing: Big data analysis technology combined with edge computing monitors the operating status of the power system under high load in real time. The AI ​​model analyzes sensor data and uses the isolation forest algorithm to identify anomalies in the system, such as line overload and abnormal equipment temperature.

[0250] Once an abnormality is detected, the system immediately triggers self-healing operations, automatically adjusts load distribution, and starts backup equipment to prevent the expansion of power outages caused by local failures. At the same time, abnormal situations are automatically reported, and the dispatch center can take manual intervention measures in a timely manner.

[0251] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise", or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article, or device. In the absence of further restrictions, the elements defined by the sentence "including a reference structure" do not exclude the existence of other identical elements in the process, method, article, or device including the elements.

[0252] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations may be made to the embodiments without departing from the principles and spirit of the invention, the scope of the invention being defined by the appended claims and their equivalents.

Claims

1. A natural disaster early warning and response method for power systems based on real-time monitoring, characterized in that: The method comprises the following steps: Sp1. Data analysis and processing: The data analysis and processing utilizes a long short-term memory network combined with a Bayesian network to analyze and process real-time monitoring data in the power system; Sp2. Generate a dispatching plan: The dispatching plan is generated based on a multi-source data fusion and optimization method combining a Kalman filter and a genetic algorithm to update the state of the power system in real time and generate an optimal dispatching plan; Sp3. Generate emergency response strategy: The emergency response strategy is generated by simulating the impact of natural disasters on the power system in real time in a virtual model through digital twin technology and adaptive dynamic programming method, and providing an emergency response strategy; Sp4. Anomaly detection and self-healing: The anomaly detection and self-healing use big data analysis combined with edge computing to perform anomaly detection and self-healing on the operating status of the power system. The big data analysis is used to extract important features from historical data and detect anomalies, and the edge computing is used to update the distributed AI model and respond to anomalies.

2. The method for natural disaster early warning and response in a power system based on real-time monitoring according to claim 1 is characterized in that: The combined algorithm of the LSTM neural network and the Bayesian network includes the following steps: Sp1, receive and process multi-dimensional time series data from the power system through LSTM neural network, including voltage, current, and frequency parameters, and generate hidden states for each time step; Sp2. In each time step, the hidden state generated by LSTM is passed as input to the Bayesian network, which calculates the risk of disaster occurrence based on historical disaster data and system state probability distribution; Sp3, based on the disaster risk probability calculated by the Bayesian network, the state weight of the LSTM network is updated to form a dynamic and continuously optimized prediction model for generating disaster warning signals in real time: LSTM neural network: Input data: time series input data is X: X={x1,x2,…,x T } where x T is the input at time t; The hidden state update formula in LSTM is as follows: i t =σ(W i ·[h t-1 ,x t ]+b i ) f t =σ(W f ·[h t-1 ,x t ]+b f ) the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t *tanh(C t ) i t 、f t , o t are the activation values ​​of the input gate, forget gate, and output gate, respectively. t is the unit state, h t is the output vector, W i , W f , W o , W C is the weight matrix, b i 、b f 、b o 、b C is bias; Bayesian network: Node representation: Bayesian network, where X = {X1, X2, ..., X n } is a set of random variables, G = (V, E) represents the directed acyclic graph of the network, V is the node set, and E is the edge set; Joint probability distribution: The joint probability distribution represented by the Bayesian network is: Pa(X i ) represents node X i By combining the time series prediction capability of LSTM with the probabilistic reasoning capability of Bayesian network, time dependency and uncertainty are handled in the prediction of complex natural disasters.

3. The method for natural disaster early warning and response of power system based on real-time monitoring according to claim 1 is characterized in that: The combined algorithm of Kalman filter and genetic algorithm comprises the following steps: Sp1, extracting power system state information from real-time data collected by multiple sensors through Kalman filter, and calculating the optimal estimate of each state variable during the filtering process; Sp2. Use genetic algorithm to globally optimize the state estimation results output by Kalman filter. The genetic algorithm evaluates the impact of different scheduling schemes on the stability of the power system through the fitness function and selects the optimal scheme; Sp3. Return the optimal dispatching scheme selected by the genetic algorithm to the Kalman filter to correct future state estimates, thereby achieving dynamic dispatch and optimization of the power system.

4. The method for natural disaster early warning and response of power system based on real-time monitoring according to claim 1 is characterized in that: The steps of combining the digital twin technology with the adaptive dynamic programming method include: Sp1. Create a virtual simulation model of the power system and dynamically calibrate the virtual model through real-time monitoring data to accurately reflect the current system status; Sp2, using adaptive dynamic programming methods to perform multiple rounds of strategy iterations in a virtual model to evaluate the effectiveness of different emergency response strategies in simulated natural disaster scenarios; Sp3. Based on the results of strategy iteration, determine and deploy the optimal emergency response strategy to the actual power system to achieve rapid response and recovery when disasters occur.

5. The method for natural disaster early warning and response in power system based on real-time monitoring according to claim 1 is characterized in that: The steps of combining big data analysis with edge computing technology include: Sp1. Through big data analysis technology, principal component analysis and autoencoder are used to reduce the dimension of historical operation data of the power system, extract key features and construct feature vectors; Sp2, deploy lightweight AI models in edge computing devices to process feature vectors obtained from sensors in real time, and use isolation forest algorithms and support vector machines to detect system anomalies; Sp3. When an abnormality is detected, the pre-trained distributed AI model is called through the edge computing device to perform local self-healing operations, and at the same time, an alarm message is sent to the central control system to achieve a multi-level response.

6. The method for natural disaster early warning and response of power system based on real-time monitoring according to claim 1 is characterized in that: The self-healing mechanism is implemented by the following steps: Sp1. Use distributed AI models to monitor the operating status of the power system in real time. When a system abnormality is detected, it quickly triggers self-healing operations, including switching power lines, starting backup power sources, and adjusting load distribution. Sp2. The optimization of self-healing operations is achieved through a reinforcement learning algorithm. The reinforcement learning model continuously optimizes future self-healing strategies based on the results of historical self-healing operations to ensure that the system stability can be quickly restored under different natural disaster conditions.

7. The device of the method for early warning and response of natural disasters in power system based on real-time monitoring according to any one of claims 1 to 6, characterized in that: include: Data acquisition module, used to collect real-time data of the power system from various sensors; Prediction analysis module, which is used to predict the impact of natural disasters on the power system based on the combined algorithm of LSTM and Bayesian network; The state update and optimization module is used to update the power system state in real time and generate a dispatch plan based on a combination of Kalman filter and genetic algorithm; The emergency response module is used to simulate disaster scenarios and generate emergency response strategies through a combination of digital twins and adaptive dynamic programming algorithms; The anomaly detection and self-healing module is used to detect anomalies in the power system and perform self-healing through a combined algorithm of big data analysis and edge computing.

8. The device for natural disaster early warning and response of power system based on real-time monitoring according to claim 7 is characterized in that: The prediction analysis module: Use a multi-layer LSTM neural network to process multi-dimensional time series data of the power system, including voltage, current, frequency, wind speed, rainfall, seismic wave environment and system parameters, and generate hidden states and predicted values ​​for multiple time steps; Based on the Bayesian network, uncertainty analysis and probability reasoning are performed on the predicted values ​​generated by the above LSTM neural network, and the probability distribution of natural disasters is calculated by combining historical natural disaster data, geographical location characteristics, and weather forecast information; The disaster probability value calculated by the Bayesian network is compared with the preset risk threshold. When the threshold is exceeded, a disaster warning signal is generated and sent. At the same time, the learning rate and weight distribution of the LSTM neural network are adjusted to improve the prediction accuracy and response speed.

9. The device for early warning and response to natural disasters in a power system based on real-time monitoring according to claim 7, characterized in that: The state update and optimization module: The Kalman filter integrates multiple sensor data in real time, including the operating status parameters of the power system and the external environment parameters, and outputs the optimal estimate of the system status; The genetic algorithm is used to perform global search and optimization on the state estimates generated by the Kalman filter. The genetic algorithm evaluates the effectiveness of different scheduling schemes based on a predefined fitness function, which takes into account system stability, energy efficiency, and disaster response time factors. Based on the optimal dispatching plan selected by the genetic algorithm, the operating mode of the power system is dynamically adjusted, including load distribution, generator start and stop, line switching, and energy storage device call-up. The Kalman filter is used to provide real-time feedback and corrections on the execution of the dispatching plan to cope with system fluctuations during disasters.

10. The device for natural disaster early warning and response of power system based on real-time monitoring according to claim 7, characterized in that: The anomaly detection and self-healing module: Through big data analysis technology, PCA and tSNE dimensionality reduction algorithms are used to process the historical operation data and real-time monitoring data of the power system, extract key feature vectors representing the health status of the power system, and construct a high-dimensional feature space model; Deploy lightweight AI models in edge computing devices, use the isolation forest algorithm and support vector machine-based anomaly detection algorithm to monitor the operating status of the power system in real time in a high-dimensional feature space model and identify potential system anomalies; When an anomaly is detected, the reinforcement learning algorithm is used to make self-healing operation decisions in the distributed AI model, optimize the self-healing strategy, and perform the following operations: switch the faulty line, start the backup power supply, adjust the load distribution, isolate the faulty area, and upload the abnormal information to the central control system for further analysis and global scheduling decisions.

Images